Skip to content

Android: 16 KB aligned injected dependencies, and a gate that resolves them - #6120

Merged
shai-almog merged 3 commits into
masterfrom
camerax-16k-alignment
Oct 28, 2026
Merged

Android: 16 KB aligned injected dependencies, and a gate that resolves them#6120
shai-almog merged 3 commits into
masterfrom
camerax-16k-alignment

Conversation

@shai-almog

@shai-almog shai-almog commented Oct 27, 2026

Copy link
Copy Markdown
Collaborator

Google Play rejected an upload for libimage_processing_util_jni.so, which comes from CameraX 1.3.4: the builder injects it for CodeScanner and the vision views. check-16k-page-alignment.py reads tracked files only, and builder-injected Gradle dependencies are downloaded at build time, so nothing checked them.

Pins: CameraX 1.4.2 (same minSdk 21, every class AndroidCameraImpl reaches by reflection is present), ML Kit, LiteRT 1.4.2 (same Interpreter/Tensor/DataType API) and ONNX Runtime 1.21.1 (Stable Diffusion floor now minSdk 24). rootbeer-lib 0.1.0 was misaligned and moves to 0.1.2; android.rootCheck lifts minSdk to 21.

Bugs the gate found: play-services-panaroma / -safenet were misspelled and never resolved; play-services-appindexing ended at 9.8.0, so android.playService.indexing now fails with a clear message.

Gate: scripts/check-android-dependency-alignment.py resolves every injected coordinate with Gradle, each group in its own configuration (pooled, highest-version-wins hid misaligned pins), and runs the ELF check. AndroidSixteenKbPageFloorTest is the offline half: measured floors, plus a ratchet that fails when a builder source names a coordinate the lister does not list. Locally: 95 groups, 72 64-bit libraries, all aligned.

The cloud builder twin (BuildDaemon, same branch name) runs this script from this repo, so this one merges first.

🤖 Generated with Claude Code

@shai-almog @claude
Android: 16 KB aligned injected dependencies, and a gate that resolve…
7a86096 
…s them

CameraX 1.3.4, injected for CodeScanner and the vision views, ships a 4 KB
aligned libimage_processing_util_jni.so, and Google Play rejects the upload.
check-16k-page-alignment.py reads tracked files only, so it never saw it.

- CameraX 1.4.2; ML Kit, LiteRT and ONNX Runtime raised to aligned releases
  (ONNX lifts the Stable Diffusion floor to minSdk 24).
- rootbeer-lib 0.1.2 (0.1.0 is misaligned); rootCheck lifts minSdk to 21.
- play-services-panorama / -safetynet were misspelled and never resolved;
  safetynet now reads its own default.
- android.playService.indexing fails with a clear message: the artifact
  ended at 9.8.0 and no version works beside a current play-services base.
- scripts/check-android-dependency-alignment.py resolves every injected
  coordinate with Gradle, each group alone, and runs the ELF check;
  AndroidSixteenKbPageFloorTest holds measured floors and ratchets the
  lister against every coordinate the builder sources name.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions

github-actions Bot commented Oct 27, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Developer Guide build artifacts are available for download from this workflow run:

Developer Guide quality checks:

  • AsciiDoc linter: No issues found (report)
  • Vale: No alerts found (report)
  • Paragraph capitalization: No paragraph capitalization issues (report)
  • LanguageTool: No grammar matches (report)
  • Image references: No unused images detected (report)

@github-actions

github-actions Bot commented Oct 27, 2026

Copy link
Copy Markdown
Contributor

Cloudflare Preview

@github-actions

github-actions Bot commented Oct 27, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

✅ Continuous Quality Report

Test & Coverage

Static Analysis

  • SpotBugs [Report archive]
    • ✅ ByteCodeTranslator: 0 findings (no issues)
    • ✅ android: 0 findings (no issues)
    • ✅ android-compat: 0 findings (no issues)
    • ✅ android-res-compiler: 0 findings (no issues)
    • ✅ backend: 0 findings (no issues)
    • ✅ backend-test: 0 findings (no issues)
    • ✅ build-engine: 0 findings (no issues)
    • ✅ build-hint-catalog: 0 findings (no issues)
    • ✅ build-hint-tools: 0 findings (no issues)
    • ✅ codenameone-gradle-plugin: 0 findings (no issues)
    • ✅ codenameone-maven-plugin: 0 findings (no issues)
    • ✅ core-unittests: 0 findings (no issues)
    • ✅ desktop-compat-runtime: 0 findings (no issues)
    • ✅ fxml-compiler: 0 findings (no issues)
    • ✅ ios: 0 findings (no issues)
    • ✅ javac: 0 findings (no issues)
    • ✅ javafx-compat: 0 findings (no issues)
    • ✅ project-model: 0 findings (no issues)
    • ✅ swing-compat: 0 findings (no issues)
  • ✅ PMD: 0 findings (no issues) [Report archive]
  • ✅ Checkstyle: 0 findings (no issues) [Report archive]

Generated automatically by the PR CI workflow.

@shai-almog

shai-almog commented Oct 27, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator Author

Compared 191 screenshots: 191 matched.

Native Android coverage

  • 📊 Line coverage: 18.90% (48250/255242 lines covered) [HTML preview] (artifact android-coverage-report, jacocoAndroidReport/html/index.html)
    • Other counters: instruction 17.64% (229182/1299291), branch 12.04% (16817/139631), complexity 9.68% (12910/133391), method 14.58% (9122/62546), class 25.14% (1774/7056)
    • Lowest covered classes
      • kotlin.collections.kotlin.collections.ArraysKt___ArraysKt – 0.00% (0/6356 lines covered)
      • kotlin.collections.unsigned.kotlin.collections.unsigned.UArraysKt___UArraysKt – 0.00% (0/2384 lines covered)
      • org.jacoco.agent.rt.internal_0e20598.asm.org.jacoco.agent.rt.internal_0e20598.asm.ClassReader – 0.00% (0/1524 lines covered)
      • org.jacoco.agent.rt.internal_0e20598.asm.org.jacoco.agent.rt.internal_0e20598.asm.MethodWriter – 0.00% (0/922 lines covered)
      • kotlinx.coroutines.channels.kotlinx.coroutines.channels.BufferedChannel – 0.00% (0/832 lines covered)
      • kotlin.sequences.kotlin.sequences.SequencesKt___SequencesKt – 0.00% (0/735 lines covered)
      • com.google.common.cache.com.google.common.cache.LocalCache$Segment – 0.00% (0/726 lines covered)
      • okio.okio.Buffer – 0.00% (0/687 lines covered)
      • com.codename1.androidcompat.rt.com.codename1.androidcompat.rt.TextXmlPullParser – 0.00% (0/623 lines covered)
      • kotlin.text.kotlin.text.StringsKt___StringsKt – 0.00% (0/623 lines covered)

✅ Native Android screenshot tests passed.

Native Android coverage

  • 📊 Line coverage: 18.90% (48250/255242 lines covered) [HTML preview] (artifact android-coverage-report, jacocoAndroidReport/html/index.html)
    • Other counters: instruction 17.64% (229182/1299291), branch 12.04% (16817/139631), complexity 9.68% (12910/133391), method 14.58% (9122/62546), class 25.14% (1774/7056)
    • Lowest covered classes
      • kotlin.collections.kotlin.collections.ArraysKt___ArraysKt – 0.00% (0/6356 lines covered)
      • kotlin.collections.unsigned.kotlin.collections.unsigned.UArraysKt___UArraysKt – 0.00% (0/2384 lines covered)
      • org.jacoco.agent.rt.internal_0e20598.asm.org.jacoco.agent.rt.internal_0e20598.asm.ClassReader – 0.00% (0/1524 lines covered)
      • org.jacoco.agent.rt.internal_0e20598.asm.org.jacoco.agent.rt.internal_0e20598.asm.MethodWriter – 0.00% (0/922 lines covered)
      • kotlinx.coroutines.channels.kotlinx.coroutines.channels.BufferedChannel – 0.00% (0/832 lines covered)
      • kotlin.sequences.kotlin.sequences.SequencesKt___SequencesKt – 0.00% (0/735 lines covered)
      • com.google.common.cache.com.google.common.cache.LocalCache$Segment – 0.00% (0/726 lines covered)
      • okio.okio.Buffer – 0.00% (0/687 lines covered)
      • com.codename1.androidcompat.rt.com.codename1.androidcompat.rt.TextXmlPullParser – 0.00% (0/623 lines covered)
      • kotlin.text.kotlin.text.StringsKt___StringsKt – 0.00% (0/623 lines covered)

Benchmark Results

Detailed Performance Metrics

Metric Duration
SIMD kernel backend scalar fallback (no native SIMD)
SIMD int-add (64K x300) java 168ms / native 114ms = 1.4x speedup
SIMD float-mul (64K x300) java 76ms / native 61ms = 1.2x speedup
SIMD kernel correctness PASS (native result == scalar reference)
Base64 payload size 8192 bytes
Base64 benchmark iterations 6000
Base64 SIMD byte path gated to scalar (CPU autovectorizes scalar; explicit SIMD not beneficial here)
Base64 CN1 encode 65.000 ms
Base64 CN1 decode 71.000 ms
Base64 native encode 356.000 ms
Base64 encode ratio (CN1/native) 0.183x (81.7% faster)
Base64 native decode 234.000 ms
Base64 decode ratio (CN1/native) 0.303x (69.7% faster)
Image encode benchmark status skipped (SIMD unsupported)

@shai-almog

shai-almog commented Oct 27, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator Author

Compared 227 screenshots: 227 matched.
✅ JavaScript-port screenshot tests passed.

@github-actions

github-actions Bot commented Oct 27, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

✅ ByteCodeTranslator Quality Report

Test & Coverage

  • ✅ Tests: 882 total, 0 failed, 55 skipped

Benchmark Results

  • Execution Time: 5912 ms

  • Hotspots (Top 20 sampled methods):

    • 8.35% java.lang.Object.hashCode (77 samples)
    • 8.13% com.codename1.tools.translator.IteratorEscape.ctorOnlyStoresParamsIntoThis (75 samples)
    • 7.38% java.util.ArrayList.indexOf (68 samples)
    • 5.42% java.lang.System.identityHashCode (50 samples)
    • 3.15% com.codename1.tools.translator.bytecodes.Invoke.findMethodUp (29 samples)
    • 2.93% com.codename1.tools.translator.BytecodeMethod.equals (27 samples)
    • 2.93% com.codename1.tools.translator.IteratorEscape.mangle (27 samples)
    • 2.71% java.lang.StringBuilder.append (25 samples)
    • 2.17% java.util.AbstractCollection.containsAll (20 samples)
    • 1.95% java.util.HashMap.hash (18 samples)
    • 1.74% com.codename1.tools.translator.Parser.cullMethods (16 samples)
    • 1.63% com.codename1.tools.translator.ByteCodeClass.hasDeclaredMethod (15 samples)
    • 1.41% com.codename1.tools.translator.ByteCodeClass.generateCCode (13 samples)
    • 1.30% java.io.FileOutputStream.writeBytes (12 samples)
    • 1.08% java.lang.String.equals (10 samples)
    • 1.08% com.codename1.tools.translator.BytecodeMethod.updateInlinableFieldDependencies (10 samples)
    • 0.98% com.codename1.tools.translator.classfile.analysis.Frame.execute (9 samples)
    • 0.98% com.codename1.tools.translator.bytecodes.Invoke.resolveDirectTarget (9 samples)
    • 0.98% com.codename1.tools.translator.DeadFieldElimination.run (9 samples)
    • 0.87% java.io.FileOutputStream.open0 (8 samples)

  • ⚠️ Coverage report not generated.

Static Analysis

  • ✅ SpotBugs: no findings (report was not generated by the build).
  • ⚠️ PMD report not generated.
  • ⚠️ Checkstyle report not generated.

Generated automatically by the PR CI workflow.

@github-actions

github-actions Bot commented Oct 27, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Flutter vs Codename One

Same application, built both ways: Flutter's AOT compile of the gallery, and the identical Dart source transpiled to Java on Codename One. Lower is better for every metric, so a ratio above 1.00x means Codename One is ahead by that factor.

android

9 interleaved rounds; host load 0.74, 1.75, 1.51

Metric Codename One Flutter Ratio Winner
Installed size 78.2 MB 95.8 MB 1.22x Codename One
Executable code 4.7 MB 20.6 MB 4.41x Codename One
Download size (zipped) 78.2 MB 95.8 MB 1.22x Codename One
Cold start to first frame on screen 469 ms 1501 ms 3.08x Codename One
Memory at rest 45.6 MB 77.5 MB 1.72x Codename One

Start-up and memory at rest are judged round by round: the ratio is the median of the per-round ratios, each from two launches made back to back, and the times and sizes shown are each side's median, so the ratio need not equal their quotient. Codename One was ahead in 9 of 9 rounds on cold start to first frame on screen; 9 of 9 rounds on memory at rest.

Compute -- the VM workloads (vm/benchmarks), run inside each app; time is the best of the app's own repetitions, higher ratio means Codename One is faster.

Workload Codename One Flutter Ratio
intArithmetic 86 ms 188 ms 2.19x
longArithmetic 64 ms 59 ms 0.92x
mathTranscendental 228 ms 345 ms 1.51x
arraySequential 31 ms 36 ms 1.16x
arrayRandom 391 ms 455 ms 1.16x
objectAllocation 67 ms 48 ms 0.72x
valueEscape 7 ms 7 ms 1.00x
hashMapChurn 147 ms 300 ms 2.04x
stringBuilding 97 ms 125 ms 1.29x
recursion 88 ms 271 ms 3.08x
quicksortBench 134 ms 154 ms 1.15x
Geometric mean 1.35x

Gate: within tolerance against scripts/flutter-bench/baseline.

Android sizes are for the arm64-v8a slice a phone is delivered, on both sides; Flutter's release apk carries three ABIs and would otherwise count its engine about three times.
Android start-up and memory come from an x86_64 emulator with a software GPU. They compare the two runtimes on one machine; they are not phone timings, and a renderer that leans on the GPU is penalised more there.
This platform's adapter has not been exercised end to end before; treat the first published run as the thing under review.

ios

9 interleaved rounds; host load 28.33, 24.77, 20.8

Metric Codename One Flutter Ratio Winner
Installed size 108.2 MB 110.5 MB 1.02x Codename One
Executable code 17.2 MB 19.1 MB 1.11x Codename One
Download size (zipped) 80.0 MB 81.9 MB 1.02x Codename One
Cold start to first frame on screen -- -- -- not measured
Memory at rest -- -- -- not measured

Compute -- the VM workloads (vm/benchmarks), run inside each app; time is the best of the app's own repetitions, higher ratio means Codename One is faster.

Not measured: iOS compute is not measured: a device build needs signed hardware, and on the simulator Flutter runs its JIT debug engine rather than AOT code. The macOS row runs the same ParparVM and Dart AOT compilers on the same Apple silicon.

Gate: within tolerance against scripts/flutter-bench/baseline. Smaller than its baseline, which does not fail the gate: Executable code -7.16% -- rebaseline to tighten. To accept the move, download this platform's baseline-<platform>.json artifact and run python3 scripts/flutter-bench/flutter_baseline.py calibrate --pr 6120 --reason "<why these moved>" baseline-ios.json, then commit the overlay it writes.

iOS start-up and memory are not measured here: Flutter cannot AOT-compile for the simulator, so a simulator run would time a JIT debug build against a release build. Sizes come from release device bundles; timings need signed hardware.
This platform's adapter has not been exercised end to end before; treat the first published run as the thing under review.

javascript

9 interleaved rounds; host load 2.33, 1.57, 0.72

Metric Codename One Flutter Ratio Winner
Installed size 104.0 MB 122.5 MB 1.18x Codename One
Executable code 12.9 MB 25.2 MB 1.95x Codename One
Download size (zipped) 76.3 MB 84.1 MB 1.10x Codename One
Cold start to first frame on screen -- -- -- not measured
Memory at rest -- -- -- not measured

Compute -- the VM workloads (vm/benchmarks), run inside each app; time is the best of the app's own repetitions, higher ratio means Codename One is faster.

Workload Codename One Flutter Ratio
intArithmetic 646 ms wrong result Flutter: wrong result
longArithmetic 4982 ms did not run Flutter: did not run
mathTranscendental 570 ms did not run Flutter: did not run
arraySequential 228 ms 82 ms 0.36x
arrayRandom 927 ms wrong result Flutter: wrong result
objectAllocation 38 ms 27 ms 0.71x
valueEscape 110 ms 7 ms 0.06x
hashMapChurn 1022 ms 245 ms 0.24x
stringBuilding 526 ms 106 ms 0.20x
recursion 508 ms 490 ms 0.96x
quicksortBench 413 ms wrong result Flutter: wrong result
Geometric mean 0.30x

Gate: within tolerance against scripts/flutter-bench/baseline; BEHIND FLUTTER on Compute (geomean of the VM workloads) (reported, not gated: a loss to Flutter on the web is currently accepted).

the JavaScript adapter needs a Chrome DevTools driver; sizes are reported and timings are not
This platform's adapter has not been exercised end to end before; treat the first published run as the thing under review.

linux

9 interleaved rounds; host load 0.22, 2.86, 2.38

Metric Codename One Flutter Ratio Winner
Installed size 111.5 MB 143.8 MB 1.29x Codename One
Executable code 14.1 MB 16.1 MB 1.15x Codename One
Download size (zipped) 80.9 MB 92.9 MB 1.15x Codename One
Cold start to first frame on screen 150 ms 220 ms 1.46x Codename One
Memory at rest 85.8 MB 225.7 MB 2.64x Codename One

Start-up and memory at rest are judged round by round: the ratio is the median of the per-round ratios, each from two launches made back to back, and the times and sizes shown are each side's median, so the ratio need not equal their quotient. Codename One was ahead in 9 of 9 rounds on cold start to first frame on screen; 9 of 9 rounds on memory at rest.

Compute -- the VM workloads (vm/benchmarks), run inside each app; time is the best of the app's own repetitions, higher ratio means Codename One is faster.

Workload Codename One Flutter Ratio
intArithmetic 74 ms 183 ms 2.47x
longArithmetic 56 ms 55 ms 0.98x
mathTranscendental 253 ms 281 ms 1.11x
arraySequential 15 ms 35 ms 2.33x
arrayRandom 235 ms 314 ms 1.34x
objectAllocation 23 ms 37 ms 1.61x
valueEscape 4 ms 7 ms 1.75x
hashMapChurn 48 ms 338 ms 7.04x
stringBuilding 32 ms 110 ms 3.44x
recursion 87 ms 258 ms 2.97x
quicksortBench 138 ms 149 ms 1.08x
Geometric mean 1.98x

Gate: within tolerance against scripts/flutter-bench/baseline.

This platform's adapter has not been exercised end to end before; treat the first published run as the thing under review.

macos

9 interleaved rounds; host load 2.02, 9.34, 13.58

Metric Codename One Flutter Ratio Winner
Installed size 124.2 MB 139.4 MB 1.12x Codename One
Executable code 33.1 MB 47.9 MB 1.45x Codename One
Download size (zipped) 85.5 MB 93.3 MB 1.09x Codename One
Cold start to first frame on screen 432 ms 619 ms 1.32x Codename One
Memory at rest 36.4 MB 53.7 MB 1.48x Codename One

Start-up and memory at rest are judged round by round: the ratio is the median of the per-round ratios, each from two launches made back to back, and the times and sizes shown are each side's median, so the ratio need not equal their quotient. Codename One was ahead in 7 of 9 rounds on cold start to first frame on screen; 9 of 9 rounds on memory at rest.

Compute -- the VM workloads (vm/benchmarks), run inside each app; time is the best of the app's own repetitions, higher ratio means Codename One is faster.

Workload Codename One Flutter Ratio
intArithmetic 106 ms 287 ms 2.71x
longArithmetic 67 ms 78 ms 1.16x
mathTranscendental 264 ms 376 ms 1.42x
arraySequential 10 ms 44 ms 4.40x
arrayRandom 253 ms 346 ms 1.37x
objectAllocation 57 ms 48 ms 0.84x
valueEscape 5 ms 15 ms 3.00x
hashMapChurn 32 ms 237 ms 7.41x
stringBuilding 23 ms 160 ms 6.96x
recursion 127 ms 307 ms 2.42x
quicksortBench 113 ms 151 ms 1.34x
Geometric mean 2.34x

Gate: within tolerance against scripts/flutter-bench/baseline. Smaller than its baseline, which does not fail the gate: Executable code -6.26% -- rebaseline to tighten. To accept the move, download this platform's baseline-<platform>.json artifact and run python3 scripts/flutter-bench/flutter_baseline.py calibrate --pr 6120 --reason "<why these moved>" baseline-macos.json, then commit the overlay it writes.

This platform's adapter has not been exercised end to end before; treat the first published run as the thing under review.

windows

9 interleaved rounds

Metric Codename One Flutter Ratio Winner
Installed size 107.8 MB 121.1 MB 1.12x Codename One
Executable code 11.4 MB 14.1 MB 1.25x Codename One
Download size (zipped) 77.9 MB 86.4 MB 1.11x Codename One
Cold start to first frame on screen 157 ms 248 ms 1.55x Codename One
Memory at rest 58.5 MB 92.7 MB 1.58x Codename One

Start-up and memory at rest are judged round by round: the ratio is the median of the per-round ratios, each from two launches made back to back, and the times and sizes shown are each side's median, so the ratio need not equal their quotient. Codename One was ahead in 9 of 9 rounds on cold start to first frame on screen; 9 of 9 rounds on memory at rest.

Compute -- the VM workloads (vm/benchmarks), run inside each app; time is the best of the app's own repetitions, higher ratio means Codename One is faster.

Workload Codename One Flutter Ratio
intArithmetic 81 ms 183 ms 2.26x
longArithmetic 64 ms 57 ms 0.89x
mathTranscendental 198 ms 267 ms 1.35x
arraySequential 10 ms 35 ms 3.50x
arrayRandom 278 ms 352 ms 1.27x
objectAllocation 24 ms 40 ms 1.67x
valueEscape 1 ms 7 ms 7.00x
hashMapChurn 47 ms 317 ms 6.75x
stringBuilding 35 ms 121 ms 3.46x
recursion 195 ms 265 ms 1.36x
quicksortBench 136 ms 148 ms 1.09x
Geometric mean 2.16x

Gate: within tolerance against scripts/flutter-bench/baseline. Smaller than its baseline, which does not fail the gate: Executable code -10.64%; Download size (zipped) -0.77% -- rebaseline to tighten. To accept the move, download this platform's baseline-<platform>.json artifact and run python3 scripts/flutter-bench/flutter_baseline.py calibrate --pr 6120 --reason "<why these moved>" baseline-windows.json, then commit the overlay it writes.

This platform's adapter has not been exercised end to end before; treat the first published run as the thing under review.

Codename One wins 26 of 26 measured metrics across 6 platform(s).

shai-almog and others added 2 commits October 27, 2026 18:23
@shai-almog @claude
ci: stage UpdateCodenameOne.jar for the dependency alignment workflow
95999ed 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@shai-almog @claude
ci: point the dependency lister at the builder sources from maven/
0400381 
The step runs in maven/, where the lister's default source path
(relative to build-engine) does not exist.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@shai-almog

shai-almog commented Oct 27, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator Author

Compared 192 screenshots: 192 matched.
✅ Native Mac screenshot tests passed.

Benchmark Results

  • VM Translation Time: 0 seconds
  • Compilation Time: 474 seconds

Detailed Performance Metrics

Metric Duration
SIMD kernel backend SSE2 (x64) / NEON (arm64) native kernels
SIMD int-add (64K x300) java 57ms / native 3ms = 19.0x speedup
SIMD float-mul (64K x300) java 53ms / native 3ms = 17.6x speedup
SIMD kernel correctness PASS (native result == scalar reference)
Base64 native bridge unavailable (CN1 + SIMD + image benchmarks only)
Base64 payload size 8192 bytes
Base64 benchmark iterations 6000
Base64 SIMD byte path active (NEON-accelerated)
Base64 CN1 encode 44.000 ms
Base64 CN1 decode 51.000 ms
Image encode benchmark iterations 100
Image createMask (SIMD off) 8.000 ms
Image createMask (SIMD on) 6.000 ms
Image createMask ratio (SIMD on/off) 0.750x (25.0% faster)
Image applyMask (SIMD off) 64.000 ms
Image applyMask (SIMD on) 78.000 ms
Image applyMask ratio (SIMD on/off) 1.219x (21.9% slower)
Image modifyAlpha (SIMD off) 60.000 ms
Image modifyAlpha (SIMD on) 38.000 ms
Image modifyAlpha ratio (SIMD on/off) 0.633x (36.7% faster)
Image modifyAlpha removeColor (SIMD off) 47.000 ms
Image modifyAlpha removeColor (SIMD on) 37.000 ms
Image modifyAlpha removeColor ratio (SIMD on/off) 0.787x (21.3% faster)

ParparVM vs HotSpot (JDK 25): macOS arm64

Runner CPU: Apple M1 (Virtual) (baseline macos-arm64)

Ratios are ParparVM / JDK 25: below 1.00x ParparVM is faster (time) or smaller (RAM). Median of 5 interleaved, paired rounds; every run's output was verified. A ratio more than 20% (time) / 25% (RAM) above its baseline in vm/selfhost/perf-baseline/ is a regression and fails; one that far below is an improvement, reported but not failed (a row whose calibration runs were noisier carries a wider tolerance), and for RAM the change must also exceed 0.05x in absolute terms. Both run unpinned on all of the runner's CPUs, with their own default thread counts.

Benchmark Cores Time RAM Status
hello (HelloCodenameOne) 3 0.84x (base 0.96x, -12.5%) 0.89x (base 0.93x, -3.5%) ok
translator (self) 3 0.64x (base 0.62x, +2.0%) 0.70x (base 0.78x, -10.1%) ok
intArithmetic 3 1.05x (base 1.03x, +2.0%) 0.12x (base 0.12x, +1.8%) ok
longArithmetic 3 1.04x (base 1.03x, +1.6%) 0.11x (base 0.07x, +63.8%) ok
mathTranscendental 3 0.96x (base 1.01x, -4.4%) 0.12x (base 0.12x, +0.9%) ok
arraySequential 3 0.43x (base 0.42x, +2.0%) 0.53x (base 0.53x, +0.4%) ok
arrayRandom 3 0.97x (base 0.99x, -1.5%) 0.32x (base 0.32x, +0.1%) ok
objectAllocation 3 1.09x (base 1.11x, -1.4%) 0.79x (base 0.72x, +8.9%) ok
valueEscape 3 0.51x (base 0.51x, +0.3%) 0.09x (base 0.09x, +0.7%) ok
hashMapChurn 3 1.07x (base 1.18x, -8.7%) 0.08x (base 0.07x, +14.4%) ok
stringBuilding 3 0.86x (base 0.84x, +2.5%) 0.54x (base 0.46x, +16.8%) ok
recursion 3 1.26x (base 1.24x, +1.9%) 0.12x (base 0.12x, +2.1%) ok
quicksort 3 1.00x (base 1.01x, -0.8%) 0.11x (base 0.11x, +0.9%) ok

Result: no regression

@shai-almog

shai-almog commented Oct 27, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator Author

Compared 184 screenshots: 184 matched.
✅ Native Apple TV (tvOS, Metal) screenshot tests passed.

@shai-almog

shai-almog commented Oct 28, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator Author

Compared 189 screenshots: 189 matched.
✅ Native iOS Metal screenshot tests passed.

Benchmark Results

  • VM Translation Time: 0 seconds
  • Compilation Time: 2324 seconds

Build and Run Timing

Metric Duration
Simulator Boot 115000 ms
Simulator Boot (Run) 1000 ms
App Install 13000 ms
App Launch 4000 ms
Test Execution 416000 ms

Detailed Performance Metrics

Metric Duration
SIMD kernel backend SSE2 (x64) / NEON (arm64) native kernels
SIMD int-add (64K x300) java 185ms / native 8ms = 23.1x speedup
SIMD float-mul (64K x300) java 107ms / native 4ms = 26.7x speedup
SIMD kernel correctness PASS (native result == scalar reference)
Base64 payload size 8192 bytes
Base64 benchmark iterations 6000
Base64 SIMD byte path active (NEON-accelerated)
Base64 CN1 encode 59.000 ms
Base64 CN1 decode 87.000 ms
Base64 native encode 638.000 ms
Base64 encode ratio (CN1/native) 0.092x (90.8% faster)
Base64 native decode 724.000 ms
Base64 decode ratio (CN1/native) 0.120x (88.0% faster)
Base64 SIMD encode 154.000 ms
Base64 encode ratio (SIMD/CN1) 2.610x (161.0% slower)
Base64 SIMD decode 53.000 ms
Base64 decode ratio (SIMD/CN1) 0.609x (39.1% faster)
Base64 encode ratio (SIMD/native) 0.241x (75.9% faster)
Base64 decode ratio (SIMD/native) 0.073x (92.7% faster)
Image encode benchmark iterations 100
Image createMask (SIMD off) 7.000 ms
Image createMask (SIMD on) 2.000 ms
Image createMask ratio (SIMD on/off) 0.286x (71.4% faster)
Image applyMask (SIMD off) 38.000 ms
Image applyMask (SIMD on) 37.000 ms
Image applyMask ratio (SIMD on/off) 0.974x (2.6% faster)
Image modifyAlpha (SIMD off) 22.000 ms
Image modifyAlpha (SIMD on) 18.000 ms
Image modifyAlpha ratio (SIMD on/off) 0.818x (18.2% faster)
Image modifyAlpha removeColor (SIMD off) 50.000 ms
Image modifyAlpha removeColor (SIMD on) 23.000 ms
Image modifyAlpha removeColor ratio (SIMD on/off) 0.460x (54.0% faster)

@shai-almog

shai-almog commented Oct 28, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator Author

Compared 249 screenshots: 249 matched.
✅ Native Apple Watch (watchOS, Core Graphics) screenshot tests passed.

@shai-almog

shai-almog commented Oct 28, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator Author

Compared 188 screenshots: 188 matched.
✅ Native Mac screenshot tests passed.

Benchmark Results

  • VM Translation Time: 0 seconds
  • Compilation Time: 687 seconds

Detailed Performance Metrics

Metric Duration
SIMD kernel backend SSE2 (x64) / NEON (arm64) native kernels
SIMD int-add (64K x300) java 70ms / native 3ms = 23.3x speedup
SIMD float-mul (64K x300) java 65ms / native 3ms = 21.6x speedup
SIMD kernel correctness PASS (native result == scalar reference)
Base64 payload size 8192 bytes
Base64 benchmark iterations 6000
Base64 SIMD byte path active (NEON-accelerated)
Base64 CN1 encode 88.000 ms
Base64 CN1 decode 70.000 ms
Base64 native encode 1008.000 ms
Base64 encode ratio (CN1/native) 0.087x (91.3% faster)
Base64 native decode 350.000 ms
Base64 decode ratio (CN1/native) 0.200x (80.0% faster)
Base64 SIMD encode 89.000 ms
Base64 encode ratio (SIMD/CN1) 1.011x (1.1% slower)
Base64 SIMD decode 70.000 ms
Base64 decode ratio (SIMD/CN1) 1.000x (0.0% slower)
Base64 encode ratio (SIMD/native) 0.088x (91.2% faster)
Base64 decode ratio (SIMD/native) 0.200x (80.0% faster)
Image encode benchmark iterations 100
Image createMask (SIMD off) 12.000 ms
Image createMask (SIMD on) 8.000 ms
Image createMask ratio (SIMD on/off) 0.667x (33.3% faster)
Image applyMask (SIMD off) 90.000 ms
Image applyMask (SIMD on) 66.000 ms
Image applyMask ratio (SIMD on/off) 0.733x (26.7% faster)
Image modifyAlpha (SIMD off) 117.000 ms
Image modifyAlpha (SIMD on) 118.000 ms
Image modifyAlpha ratio (SIMD on/off) 1.009x (0.9% slower)
Image modifyAlpha removeColor (SIMD off) 82.000 ms
Image modifyAlpha removeColor (SIMD on) 61.000 ms
Image modifyAlpha removeColor ratio (SIMD on/off) 0.744x (25.6% faster)

@shai-almog

shai-almog commented Oct 28, 2026 •
edited
Loading

Copy link
Copy Markdown
Collaborator Author

Compared 189 screenshots: 189 matched.
✅ Native iOS Metal screenshot tests passed.

Benchmark Results

  • VM Translation Time: 0 seconds
  • Compilation Time: 1843 seconds

Build and Run Timing

Metric Duration
Simulator Boot 109000 ms
Simulator Boot (Run) 1000 ms
App Install 22000 ms
App Launch 5000 ms
Test Execution 508000 ms

Detailed Performance Metrics

Metric Duration
SIMD kernel backend SSE2 (x64) / NEON (arm64) native kernels
SIMD int-add (64K x300) java 116ms / native 3ms = 38.6x speedup
SIMD float-mul (64K x300) java 60ms / native 2ms = 30.0x speedup
SIMD kernel correctness PASS (native result == scalar reference)
Base64 payload size 8192 bytes
Base64 benchmark iterations 6000
Base64 SIMD byte path active (NEON-accelerated)
Base64 CN1 encode 341.000 ms
Base64 CN1 decode 99.000 ms
Base64 native encode 547.000 ms
Base64 encode ratio (CN1/native) 0.623x (37.7% faster)
Base64 native decode 743.000 ms
Base64 decode ratio (CN1/native) 0.133x (86.7% faster)
Base64 SIMD encode 56.000 ms
Base64 encode ratio (SIMD/CN1) 0.164x (83.6% faster)
Base64 SIMD decode 106.000 ms
Base64 decode ratio (SIMD/CN1) 1.071x (7.1% slower)
Base64 encode ratio (SIMD/native) 0.102x (89.8% faster)
Base64 decode ratio (SIMD/native) 0.143x (85.7% faster)
Image encode benchmark iterations 100
Image createMask (SIMD off) 6.000 ms
Image createMask (SIMD on) 2.000 ms
Image createMask ratio (SIMD on/off) 0.333x (66.7% faster)
Image applyMask (SIMD off) 155.000 ms
Image applyMask (SIMD on) 28.000 ms
Image applyMask ratio (SIMD on/off) 0.181x (81.9% faster)
Image modifyAlpha (SIMD off) 113.000 ms
Image modifyAlpha (SIMD on) 28.000 ms
Image modifyAlpha ratio (SIMD on/off) 0.248x (75.2% faster)
Image modifyAlpha removeColor (SIMD off) 161.000 ms
Image modifyAlpha removeColor (SIMD on) 60.000 ms
Image modifyAlpha removeColor ratio (SIMD on/off) 0.373x (62.7% faster)

@shai-almog
shai-almog merged commit cb9b544 into master Oct 28, 2026
89 of 90 checks passed
@shai-almog
shai-almog deleted the camerax-16k-alignment branch October 28, 2026 06:44
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@shai-almog